How To Build a Custom GPT for Body Shop Reviews - Episode 11 Hero Group AI Labs

Thomas Zoebelein builds a custom GPT that writes responses to Google and Facebook reviews for a collision repair shop. The GPT follows a system prompt that keeps replies warm, avoids legal admissions, and sorts issues by type. He tests it on positive and negative reviews, then shows how to customize the bot's name, icon, and sharing settings.
Chapters
If you use ChatGPT a lot, let me ask you something. Are you actually using custom GPTs? Because actually I wasn't. I didn't even know that they were a thing until my wife pointed them out to me. Meanwhile, I'm over here building all these projects when what I really needed were small task-specific GPTs for the stuff that I do every week, like these episodes. So today I want to show you what a custom GPT is, how to think about them, and how that you— how you can build one that you can actually use. Hey again, it's Tom and I'm Hi, back at it here at the Hero Group AI Labs. Listen, it's been a few weeks since I've been back here in the lab. We're getting really close to launching our first product, Reclaimer, that's going to be debuting here at the Carolina Collision Conference. And I've been on the road speaking. So if you follow our content and our newsletter, thanks for sticking with us through this gap. We had to kind of repurpose some older content. But today, like I mentioned, I'm going to show you how to build a custom GPT. The way I think about these is pretty simple. They're mini task-oriented robots that basically are like a set of instructions, kind of like a prompt, but built to handle a specific job over and over again. For example, I have one for my podcast. I can drop in a transcript and it will timecode it, tell me what it thinks the episode is really about, suggest titles, write the copy that I need to upload on the episode in my, um, on YouTube and onto my podcast hosting. I have another one for these AI Labs episodes. It helps me write the title, the YouTube upload copy, so that instead of it taking hours, it just takes a couple of minutes. My wife's actually the one who turned me on to all of this. She works for a credit union with multiple branches, and she has custom GPTs that know the whole company, understand the branch-specific needs, and they'll help with tasks tied to different locations. So I thought to build a simple one today, one that will give you a clear example and so that you can actually— that you can actually use would be a review response GPT. When I owned my marketing agency, we did review responses for people. We didn't use custom GPTs then because they— we didn't really have them. And I actually, I didn't even know about them, but we would handwrite all the responses to the reviews. So I figured this would be a good opportunity for me to build something that you can use. So we'll, we'll build you a ChatGPT review responder. So I'm going to build it. I'm going to show you the prompt that I use, and then I'll walk you through how you can tweak it for your own shop. The idea is pretty simple. You're just going to drop in a Google review and it'll give you back a response that you can use as a starting point. But the bigger lesson today is not just this one GPT, it's the best practices behind building custom GPTs that are actually useful. And like everything in AI, this stuff can change. But as of right now, these are the best practices for the tools that we have today. So let's get started. All right. So I'm going to give you kind of a behind the scenes look here. So this is actually my GPT that I used to build this episode. And I moved some stuff out of the screen here. Um, so this is, uh, like I said, this is the— this is my GPT. It's called the AI Labs Producer. So I use this when I am building an episode. So I'll tell it, you know, I'm doing AI Labs today, I'm thinking about doing about this, and then it helps me kind of organize my thoughts. And then what I do on mine is I, um, I'll drop in, uh, like a whole section. I'll just, you know, like stream of consciousness, just drop in. I want to cover this. I want to say this, this, this, and this. And then it'll help me build out the episode. And so here it is right here. So this is like the stream, you know, for the, you know, I want to say this, I want to do this, I want to mention this, this, this, and this. So then it'll help me write the open, the main intro, and then give me a couple of options. And then when I land on something that I want, it builds out a framework, what I want to cover on the episode, and then it will give me some suggestions for the— what to put on the, um, uh, you know, like, like what to cover on the episode, like what should go in the prompt. And then, um, it writes my YouTube description, it'll write my YouTube title options, episode name options, and they all fit like a format that I have, so I don't have to keep going back and thinking about what's the format that we use for this. And then suggested chapter outlines, and then some, some ideas, some call to actions and things like that, and then a closing. So that's this ChatGPT, that's this GPT. And so this is it here under GPTs. There's this thing called Explore GPTs, so you can kind of click here, um, and look for pre-built GPTs. Um, there's tons of them. There's productivity ones, writing, research and analysis, programming, education, lifestyle, so on and so on. And you can see, so these are GPTs that people have built. These are some built by ChatGPT. There's some image generating ones, some writing ones, productivity ones, um, you know, uh, some can help you with PowerPoints, photorealistic image generators, things like that. Research and analysis, probably very helpful to a body shop. So anyway, uh, that's that. So, so then we go to my GPTs, and these are the ones that I've created. I have a Labs Producer, then I have one for my podcast. And so we're going to go back here and we're going to create a new GPT. Okay, so this is the, you know, what would you like me to make? So I already have this right now. Let me just copy it. All right, so let me grab the, um, the prompt here.
Uh, where is it?
Um, you're, you're a, uh, you are a review response assistant for the Collision Repair Shop. Your job is to help the Shoprite professional human brand save responses to customer reviews on platforms like Google and Facebook. Your responses should sound warm, calm, direct, and professional. They should never sound robotic, overly corporate, defensive, or overly promotional. You're writing for a collision repair business, so you should understand the kinds of issues customers often mention, including repair quality, communication delays, parts availability, insurance involvement, scheduling, delivery timing, overall customer service. When responding to reviews, follow these rules. Now, this is all the stuff you can change, right? So you can change how you want to respond, but, you know, keep responses concise and easy to read. Always thank the customer when appropriate. For positive reviews, sound appreciative and specific. For mixed or negative reviews, acknowledge the concern without arguing. Do not admit legal fault, liability, or wrongdoing. Do not blame the customer, insurer, vendor, or parts supplier in a public reply. When the issue is sensitive, invite the customer to continue the conversation offline. Protect the shop's reputation while still sounding human. Do not use cheesy marketing language. Do not over-apologize. Do make promises the shop may not be able to keep. If a review contains a serious complaint, recommend an internal follow-up. So, um, for each review, return your answer in the exact format: sentiment, issue type, recommended approach, public response draft, internal follow-up note. Uh, issue type should be one of the following: positive communication, repair quality, delay or timing, price, cost, blah, blah, blah. Recommended approach should be, uh, one of the following: public response only, public response Plus offline follow-up. Escalate internally before responding. Public responses should usually be between 2 to 5 sentences. If the review is very short or vague, write a response that is polite and measured without making assumptions. If the review is emotional or hostile, still stay calm and professional. If the review is positive, reinforce application appreciation and trust. And then here's some brand tuning stuff in here, so it will help you Uh, sorry, I keep clicking this. Uh, it'll help you to, um, you know, make it, uh, brand specific. And then, uh, knowledge context to add, so you can add shop-specific context like all this stuff here. So I'm gonna, um, so let me just grab all this and I'm gonna drop it over there into my Prompt Builder, and I'm going to tweak a few things here. This is just straight out of GPT, the first go-around. So let me go to, uh, GPT's, create a new one. I'm gonna drop in all this copy here.
And then we'll make some changes. Let's configure it first.
So this is gonna be the Google—
views—
responder. Description, um, instructions. What does this GPT do? How does it behave? What should it do, avoid doing? Okay, so now I'm going to copy this again.
So, um, All right, I'm gonna put that in here, but I'm gonna tweak a bunch of stuff.
All right, um, here are a few responses, blah blah. Follow up is always thank the customer when appropriate. Uh, I'd like to, um, always mention the customer's name where possible. Uh, for positive reviews, sound appreciative and specific. Um, if the review mentions their car brand Try and work that into the review. R-E-V-I-E-W. Example: Glad we were able to help you with your Toyota For mixed or negative reviews, acknowledge the concern without arguing. Do not admit legal fault, wrongdoing. Do not blame the customer. If the issue's sensitive, invite the customer to continue the conversation offline by contacting the shop and asking to speak with the owner. Protect the shop's reputation while still sounding human. Do not use cheesy marketing language. Do not over-apologize. Do not, do not make promises the shop may not be able to keep. If a review contains a serious complaint, recommend an internal follow-up. Uh, for every review, return your answer in this exact format: sentiment, issue type. So this is how you'll receive your response back. Issue type should be one of the following: Recommended approach should be one of the following: public response only, public response plus offline follow-up, escalate internally before responding. This is optional stuff. If you work for somebody and you want to have this thing flag it, you know, like at my agency, we would— if we got anything under a 3-star, we would reach out to the customer with with a pre-written review and ask them, is this what you want us to put on there? And so this is for that reason here. You can get rid of it if you're the owner and you're responding to your own reviews. Public responses should only be between 2 and 5 sentences. If the review is very short or vague, write a response that's polite and measured. If it's emotional, hostile, stay calm and professional. It's positive reinforce.
Okay. Get rid of this.
This was for me. So, uh, if you're building this for yourself, so I'm going to call this Acme Autobody. It's my fake auto body shop, uh, and we're in Baltimore. Preferred tone. Like, you could have these things be funny, fun, and upbeat. Let's just see what that's, uh, that does, whether they want short or medium length responses. Um. You can't actually remove negative reviews, so— want every negative review moved offline. Yeah, actually, that's— sorry, I thought it said remove. Um, yeah, um, just—
no, actually, I'm gonna get rid of that.
Um, Common shop issues to account for. Brand phrase is a dislike. Example: We're a collision repair shop focused on clear communication, quality repairs, and professional customer experience. Okay, all right, um. So there's my instructions. Conversation starters. I think that for this one we can go with instant I think that's all I need to do. So you can share this with your coworkers.
Uh, UGPT. All right, now—
Edit it.
How do I edit it?
Edit GPT.
There we go.
Okay, now, um, now we've got our GPT built and now we're going to test it. So I'm just going to come grab some— we're going to test the review. So here's one here, let's put it in our little chat thing here. So boom, there we have it. Thank you for the great review. We're glad to hear your truck came out looking great and the process felt smooth and easy. Keeping you updated along the way is a big part of what we aim for, so it's great to know it stood out. We really appreciate the recommendation. So you just copy that and just put that in your Google and there's no follow-up needed. All right, let's do another one. Mixed. Let's see how it handles a mixed review. And just for fun, I'm just gonna put in here, um, repair look good on my F-150. Let's see what it does with that. Thank you for feedback. We're glad to hear your F-150 repair looks good. We're sorry to hear the communication and timeline didn't meet expectations. That's something we take seriously. Uh, I also appreciate you mentioning your experience at the front desk, as consistency matters to us. If you're open to it, please reach out to and ask for the owner so we can talk through your experience and work to improve. So that's pretty Pretty good handling of that one. Um, let's jump to a negative one here. Let's just see what happens with our GPTs if we need to tweak anything.
Boom.
Uh, sorry to hear about your experience and understand how frustrating that may have been. Delays and lack of clear updates are not the experience we aim to provide, and we also want to make sure all your questions are fully answered. Please reach out and ask us for— ask for the owners so we can go over everything with you and provide you clarity you deserve. We appreciate you bringing this to our attention. Um, all right, and then let's see what it does with a vague negative review. Okay, we're sorry to see this, see this, and that your experience didn't meet expectations. That's not the level of service we aim to provide. If you're willing, please reach out and ask for the owner so we can learn more about what happened and see how we can make it right. So again, these are good, you know, like what this is going to do is just save you from every time having to go start a new chat and say you know, write a Google review, you know, just all this internal stuff, right? So it's just a, just a Google Reviews responder. There's one last thing I want to show you that you can do on here. You can actually make a little icon for it so that it stands out right here. By the way, you can change the model too, right? If you want to do Pro or Thinking or, um, yeah, I'm going to change it to Pro right now. And, um, and by the way, you can just upload a thumbnail for it here if you want. Um, so I'm going to update this. All right guys, so that's it. That's a pretty fast one today, but I think it's a super useful tool. If you start thinking about these GPTs as like little robot things that understand context, and you know, some of the repetitive tasks like writing specific emails, maybe insurance adjuster emails, or you know, maybe it's updates to a bookkeeper, and we just want to just drop in some unstructured data and it will you know, go and perform a task that you do over and over and over again. Maybe it's, um, you know, dealing with a vendor, uh, repeatedly, your tow company, whatever. I'm sure you can think of little things that you do in ChatGPT, but this is just a way to think about it differently. Just building these little mini tools where all you got to do is just drop in the thing and it does it. You copy-paste it and off you go. So pretty simple. Hopefully that helped, and I will see you guys on the next AI Labs.
More from Thomas

Sal and Vinny Grill me about Insurance company estimate denials at body shops

Customer Service as a Competitive Advantage - Dan Bernier - Moay Collision Episode 10

How to Build a Daily Competitor Report With Claude Episode 13 Hero Group AI Labs

These 3 Free AI Tools Can Help Collision Shops Hire, Sell, and Grow Hero Labs Episode 12






